4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/acct.h> /* acct_auto_close_mnt */
20 #include <linux/init.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include <linux/bootmem.h>
30 static unsigned int m_hash_mask __read_mostly;
31 static unsigned int m_hash_shift __read_mostly;
32 static unsigned int mp_hash_mask __read_mostly;
33 static unsigned int mp_hash_shift __read_mostly;
35 static __initdata unsigned long mhash_entries;
36 static int __init set_mhash_entries(char *str)
40 mhash_entries = simple_strtoul(str, &str, 0);
43 __setup("mhash_entries=", set_mhash_entries);
45 static __initdata unsigned long mphash_entries;
46 static int __init set_mphash_entries(char *str)
50 mphash_entries = simple_strtoul(str, &str, 0);
53 __setup("mphash_entries=", set_mphash_entries);
56 static DEFINE_IDA(mnt_id_ida);
57 static DEFINE_IDA(mnt_group_ida);
58 static DEFINE_SPINLOCK(mnt_id_lock);
59 static int mnt_id_start = 0;
60 static int mnt_group_start = 1;
62 static struct hlist_head *mount_hashtable __read_mostly;
63 static struct hlist_head *mountpoint_hashtable __read_mostly;
64 static struct kmem_cache *mnt_cache __read_mostly;
65 static DECLARE_RWSEM(namespace_sem);
68 struct kobject *fs_kobj;
69 EXPORT_SYMBOL_GPL(fs_kobj);
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
81 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
83 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
85 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
89 static inline struct hlist_head *mp_hash(struct dentry *dentry)
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
100 static int mnt_alloc_id(struct mount *mnt)
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
106 spin_lock(&mnt_id_lock);
107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
109 mnt_id_start = mnt->mnt_id + 1;
110 spin_unlock(&mnt_id_lock);
117 static void mnt_free_id(struct mount *mnt)
119 int id = mnt->mnt_id;
120 spin_lock(&mnt_id_lock);
121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
124 spin_unlock(&mnt_id_lock);
128 * Allocate a new peer group ID
130 * mnt_group_ida is protected by namespace_sem
132 static int mnt_alloc_group_id(struct mount *mnt)
136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
139 res = ida_get_new_above(&mnt_group_ida,
143 mnt_group_start = mnt->mnt_group_id + 1;
149 * Release a peer group ID
151 void mnt_release_group_id(struct mount *mnt)
153 int id = mnt->mnt_group_id;
154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
157 mnt->mnt_group_id = 0;
161 * vfsmount lock must be held for read
163 static inline void mnt_add_count(struct mount *mnt, int n)
166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
175 * vfsmount lock must be held for write
177 unsigned int mnt_get_count(struct mount *mnt)
180 unsigned int count = 0;
183 for_each_possible_cpu(cpu) {
184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
189 return mnt->mnt_count;
193 static struct mount *alloc_vfsmnt(const char *name)
195 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
199 err = mnt_alloc_id(mnt);
204 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
205 if (!mnt->mnt_devname)
210 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
212 goto out_free_devname;
214 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
217 mnt->mnt_writers = 0;
220 INIT_HLIST_NODE(&mnt->mnt_hash);
221 INIT_LIST_HEAD(&mnt->mnt_child);
222 INIT_LIST_HEAD(&mnt->mnt_mounts);
223 INIT_LIST_HEAD(&mnt->mnt_list);
224 INIT_LIST_HEAD(&mnt->mnt_expire);
225 INIT_LIST_HEAD(&mnt->mnt_share);
226 INIT_LIST_HEAD(&mnt->mnt_slave_list);
227 INIT_LIST_HEAD(&mnt->mnt_slave);
228 #ifdef CONFIG_FSNOTIFY
229 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
236 kfree(mnt->mnt_devname);
241 kmem_cache_free(mnt_cache, mnt);
246 * Most r/o checks on a fs are for operations that take
247 * discrete amounts of time, like a write() or unlink().
248 * We must keep track of when those operations start
249 * (for permission checks) and when they end, so that
250 * we can determine when writes are able to occur to
254 * __mnt_is_readonly: check whether a mount is read-only
255 * @mnt: the mount to check for its write status
257 * This shouldn't be used directly ouside of the VFS.
258 * It does not guarantee that the filesystem will stay
259 * r/w, just that it is right *now*. This can not and
260 * should not be used in place of IS_RDONLY(inode).
261 * mnt_want/drop_write() will _keep_ the filesystem
264 int __mnt_is_readonly(struct vfsmount *mnt)
266 if (mnt->mnt_flags & MNT_READONLY)
268 if (mnt->mnt_sb->s_flags & MS_RDONLY)
272 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
274 static inline void mnt_inc_writers(struct mount *mnt)
277 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
283 static inline void mnt_dec_writers(struct mount *mnt)
286 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
292 static unsigned int mnt_get_writers(struct mount *mnt)
295 unsigned int count = 0;
298 for_each_possible_cpu(cpu) {
299 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
304 return mnt->mnt_writers;
308 static int mnt_is_readonly(struct vfsmount *mnt)
310 if (mnt->mnt_sb->s_readonly_remount)
312 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
314 return __mnt_is_readonly(mnt);
318 * Most r/o & frozen checks on a fs are for operations that take discrete
319 * amounts of time, like a write() or unlink(). We must keep track of when
320 * those operations start (for permission checks) and when they end, so that we
321 * can determine when writes are able to occur to a filesystem.
324 * __mnt_want_write - get write access to a mount without freeze protection
325 * @m: the mount on which to take a write
327 * This tells the low-level filesystem that a write is about to be performed to
328 * it, and makes sure that writes are allowed (mnt it read-write) before
329 * returning success. This operation does not protect against filesystem being
330 * frozen. When the write operation is finished, __mnt_drop_write() must be
331 * called. This is effectively a refcount.
333 int __mnt_want_write(struct vfsmount *m)
335 struct mount *mnt = real_mount(m);
339 mnt_inc_writers(mnt);
341 * The store to mnt_inc_writers must be visible before we pass
342 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
343 * incremented count after it has set MNT_WRITE_HOLD.
346 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
349 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
350 * be set to match its requirements. So we must not load that until
351 * MNT_WRITE_HOLD is cleared.
354 if (mnt_is_readonly(m)) {
355 mnt_dec_writers(mnt);
364 * mnt_want_write - get write access to a mount
365 * @m: the mount on which to take a write
367 * This tells the low-level filesystem that a write is about to be performed to
368 * it, and makes sure that writes are allowed (mount is read-write, filesystem
369 * is not frozen) before returning success. When the write operation is
370 * finished, mnt_drop_write() must be called. This is effectively a refcount.
372 int mnt_want_write(struct vfsmount *m)
376 sb_start_write(m->mnt_sb);
377 ret = __mnt_want_write(m);
379 sb_end_write(m->mnt_sb);
382 EXPORT_SYMBOL_GPL(mnt_want_write);
385 * mnt_clone_write - get write access to a mount
386 * @mnt: the mount on which to take a write
388 * This is effectively like mnt_want_write, except
389 * it must only be used to take an extra write reference
390 * on a mountpoint that we already know has a write reference
391 * on it. This allows some optimisation.
393 * After finished, mnt_drop_write must be called as usual to
394 * drop the reference.
396 int mnt_clone_write(struct vfsmount *mnt)
398 /* superblock may be r/o */
399 if (__mnt_is_readonly(mnt))
402 mnt_inc_writers(real_mount(mnt));
406 EXPORT_SYMBOL_GPL(mnt_clone_write);
409 * __mnt_want_write_file - get write access to a file's mount
410 * @file: the file who's mount on which to take a write
412 * This is like __mnt_want_write, but it takes a file and can
413 * do some optimisations if the file is open for write already
415 int __mnt_want_write_file(struct file *file)
417 if (!(file->f_mode & FMODE_WRITER))
418 return __mnt_want_write(file->f_path.mnt);
420 return mnt_clone_write(file->f_path.mnt);
424 * mnt_want_write_file - get write access to a file's mount
425 * @file: the file who's mount on which to take a write
427 * This is like mnt_want_write, but it takes a file and can
428 * do some optimisations if the file is open for write already
430 int mnt_want_write_file(struct file *file)
434 sb_start_write(file->f_path.mnt->mnt_sb);
435 ret = __mnt_want_write_file(file);
437 sb_end_write(file->f_path.mnt->mnt_sb);
440 EXPORT_SYMBOL_GPL(mnt_want_write_file);
443 * __mnt_drop_write - give up write access to a mount
444 * @mnt: the mount on which to give up write access
446 * Tells the low-level filesystem that we are done
447 * performing writes to it. Must be matched with
448 * __mnt_want_write() call above.
450 void __mnt_drop_write(struct vfsmount *mnt)
453 mnt_dec_writers(real_mount(mnt));
458 * mnt_drop_write - give up write access to a mount
459 * @mnt: the mount on which to give up write access
461 * Tells the low-level filesystem that we are done performing writes to it and
462 * also allows filesystem to be frozen again. Must be matched with
463 * mnt_want_write() call above.
465 void mnt_drop_write(struct vfsmount *mnt)
467 __mnt_drop_write(mnt);
468 sb_end_write(mnt->mnt_sb);
470 EXPORT_SYMBOL_GPL(mnt_drop_write);
472 void __mnt_drop_write_file(struct file *file)
474 __mnt_drop_write(file->f_path.mnt);
477 void mnt_drop_write_file(struct file *file)
479 mnt_drop_write(file->f_path.mnt);
481 EXPORT_SYMBOL(mnt_drop_write_file);
483 static int mnt_make_readonly(struct mount *mnt)
488 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
490 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
491 * should be visible before we do.
496 * With writers on hold, if this value is zero, then there are
497 * definitely no active writers (although held writers may subsequently
498 * increment the count, they'll have to wait, and decrement it after
499 * seeing MNT_READONLY).
501 * It is OK to have counter incremented on one CPU and decremented on
502 * another: the sum will add up correctly. The danger would be when we
503 * sum up each counter, if we read a counter before it is incremented,
504 * but then read another CPU's count which it has been subsequently
505 * decremented from -- we would see more decrements than we should.
506 * MNT_WRITE_HOLD protects against this scenario, because
507 * mnt_want_write first increments count, then smp_mb, then spins on
508 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
509 * we're counting up here.
511 if (mnt_get_writers(mnt) > 0)
514 mnt->mnt.mnt_flags |= MNT_READONLY;
516 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
517 * that become unheld will see MNT_READONLY.
520 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
525 static void __mnt_unmake_readonly(struct mount *mnt)
528 mnt->mnt.mnt_flags &= ~MNT_READONLY;
532 int sb_prepare_remount_readonly(struct super_block *sb)
537 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
538 if (atomic_long_read(&sb->s_remove_count))
542 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
543 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
544 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
546 if (mnt_get_writers(mnt) > 0) {
552 if (!err && atomic_long_read(&sb->s_remove_count))
556 sb->s_readonly_remount = 1;
559 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
560 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
561 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
568 static void free_vfsmnt(struct mount *mnt)
570 kfree(mnt->mnt_devname);
572 free_percpu(mnt->mnt_pcp);
574 kmem_cache_free(mnt_cache, mnt);
577 static void delayed_free_vfsmnt(struct rcu_head *head)
579 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
582 /* call under rcu_read_lock */
583 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
586 if (read_seqretry(&mount_lock, seq))
590 mnt = real_mount(bastard);
591 mnt_add_count(mnt, 1);
592 if (likely(!read_seqretry(&mount_lock, seq)))
594 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
595 mnt_add_count(mnt, -1);
605 * find the first mount at @dentry on vfsmount @mnt.
606 * call under rcu_read_lock()
608 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
610 struct hlist_head *head = m_hash(mnt, dentry);
613 hlist_for_each_entry_rcu(p, head, mnt_hash)
614 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
620 * find the last mount at @dentry on vfsmount @mnt.
621 * mount_lock must be held.
623 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
625 struct mount *p, *res;
626 res = p = __lookup_mnt(mnt, dentry);
629 hlist_for_each_entry_continue(p, mnt_hash) {
630 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
639 * lookup_mnt - Return the first child mount mounted at path
641 * "First" means first mounted chronologically. If you create the
644 * mount /dev/sda1 /mnt
645 * mount /dev/sda2 /mnt
646 * mount /dev/sda3 /mnt
648 * Then lookup_mnt() on the base /mnt dentry in the root mount will
649 * return successively the root dentry and vfsmount of /dev/sda1, then
650 * /dev/sda2, then /dev/sda3, then NULL.
652 * lookup_mnt takes a reference to the found vfsmount.
654 struct vfsmount *lookup_mnt(struct path *path)
656 struct mount *child_mnt;
662 seq = read_seqbegin(&mount_lock);
663 child_mnt = __lookup_mnt(path->mnt, path->dentry);
664 m = child_mnt ? &child_mnt->mnt : NULL;
665 } while (!legitimize_mnt(m, seq));
670 static struct mountpoint *new_mountpoint(struct dentry *dentry)
672 struct hlist_head *chain = mp_hash(dentry);
673 struct mountpoint *mp;
676 hlist_for_each_entry(mp, chain, m_hash) {
677 if (mp->m_dentry == dentry) {
678 /* might be worth a WARN_ON() */
679 if (d_unlinked(dentry))
680 return ERR_PTR(-ENOENT);
686 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
688 return ERR_PTR(-ENOMEM);
690 ret = d_set_mounted(dentry);
696 mp->m_dentry = dentry;
698 hlist_add_head(&mp->m_hash, chain);
702 static void put_mountpoint(struct mountpoint *mp)
704 if (!--mp->m_count) {
705 struct dentry *dentry = mp->m_dentry;
706 spin_lock(&dentry->d_lock);
707 dentry->d_flags &= ~DCACHE_MOUNTED;
708 spin_unlock(&dentry->d_lock);
709 hlist_del(&mp->m_hash);
714 static inline int check_mnt(struct mount *mnt)
716 return mnt->mnt_ns == current->nsproxy->mnt_ns;
720 * vfsmount lock must be held for write
722 static void touch_mnt_namespace(struct mnt_namespace *ns)
726 wake_up_interruptible(&ns->poll);
731 * vfsmount lock must be held for write
733 static void __touch_mnt_namespace(struct mnt_namespace *ns)
735 if (ns && ns->event != event) {
737 wake_up_interruptible(&ns->poll);
742 * vfsmount lock must be held for write
744 static void detach_mnt(struct mount *mnt, struct path *old_path)
746 old_path->dentry = mnt->mnt_mountpoint;
747 old_path->mnt = &mnt->mnt_parent->mnt;
748 mnt->mnt_parent = mnt;
749 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
750 list_del_init(&mnt->mnt_child);
751 hlist_del_init_rcu(&mnt->mnt_hash);
752 put_mountpoint(mnt->mnt_mp);
757 * vfsmount lock must be held for write
759 void mnt_set_mountpoint(struct mount *mnt,
760 struct mountpoint *mp,
761 struct mount *child_mnt)
764 mnt_add_count(mnt, 1); /* essentially, that's mntget */
765 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
766 child_mnt->mnt_parent = mnt;
767 child_mnt->mnt_mp = mp;
771 * vfsmount lock must be held for write
773 static void attach_mnt(struct mount *mnt,
774 struct mount *parent,
775 struct mountpoint *mp)
777 mnt_set_mountpoint(parent, mp, mnt);
778 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
779 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
783 * vfsmount lock must be held for write
785 static void commit_tree(struct mount *mnt, struct mount *shadows)
787 struct mount *parent = mnt->mnt_parent;
790 struct mnt_namespace *n = parent->mnt_ns;
792 BUG_ON(parent == mnt);
794 list_add_tail(&head, &mnt->mnt_list);
795 list_for_each_entry(m, &head, mnt_list)
798 list_splice(&head, n->list.prev);
801 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
803 hlist_add_head_rcu(&mnt->mnt_hash,
804 m_hash(&parent->mnt, mnt->mnt_mountpoint));
805 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
806 touch_mnt_namespace(n);
809 static struct mount *next_mnt(struct mount *p, struct mount *root)
811 struct list_head *next = p->mnt_mounts.next;
812 if (next == &p->mnt_mounts) {
816 next = p->mnt_child.next;
817 if (next != &p->mnt_parent->mnt_mounts)
822 return list_entry(next, struct mount, mnt_child);
825 static struct mount *skip_mnt_tree(struct mount *p)
827 struct list_head *prev = p->mnt_mounts.prev;
828 while (prev != &p->mnt_mounts) {
829 p = list_entry(prev, struct mount, mnt_child);
830 prev = p->mnt_mounts.prev;
836 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
842 return ERR_PTR(-ENODEV);
844 mnt = alloc_vfsmnt(name);
846 return ERR_PTR(-ENOMEM);
848 if (flags & MS_KERNMOUNT)
849 mnt->mnt.mnt_flags = MNT_INTERNAL;
851 root = mount_fs(type, flags, name, data);
855 return ERR_CAST(root);
858 mnt->mnt.mnt_root = root;
859 mnt->mnt.mnt_sb = root->d_sb;
860 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
861 mnt->mnt_parent = mnt;
863 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
867 EXPORT_SYMBOL_GPL(vfs_kern_mount);
869 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
872 struct super_block *sb = old->mnt.mnt_sb;
876 mnt = alloc_vfsmnt(old->mnt_devname);
878 return ERR_PTR(-ENOMEM);
880 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
881 mnt->mnt_group_id = 0; /* not a peer of original */
883 mnt->mnt_group_id = old->mnt_group_id;
885 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
886 err = mnt_alloc_group_id(mnt);
891 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
892 /* Don't allow unprivileged users to change mount flags */
893 if (flag & CL_UNPRIVILEGED) {
894 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
896 if (mnt->mnt.mnt_flags & MNT_READONLY)
897 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
899 if (mnt->mnt.mnt_flags & MNT_NODEV)
900 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
902 if (mnt->mnt.mnt_flags & MNT_NOSUID)
903 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
905 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
906 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
909 /* Don't allow unprivileged users to reveal what is under a mount */
910 if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
911 mnt->mnt.mnt_flags |= MNT_LOCKED;
913 atomic_inc(&sb->s_active);
914 mnt->mnt.mnt_sb = sb;
915 mnt->mnt.mnt_root = dget(root);
916 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
917 mnt->mnt_parent = mnt;
919 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
922 if ((flag & CL_SLAVE) ||
923 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
924 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
925 mnt->mnt_master = old;
926 CLEAR_MNT_SHARED(mnt);
927 } else if (!(flag & CL_PRIVATE)) {
928 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
929 list_add(&mnt->mnt_share, &old->mnt_share);
930 if (IS_MNT_SLAVE(old))
931 list_add(&mnt->mnt_slave, &old->mnt_slave);
932 mnt->mnt_master = old->mnt_master;
934 if (flag & CL_MAKE_SHARED)
937 /* stick the duplicate mount on the same expiry list
938 * as the original if that was on one */
939 if (flag & CL_EXPIRE) {
940 if (!list_empty(&old->mnt_expire))
941 list_add(&mnt->mnt_expire, &old->mnt_expire);
952 static void mntput_no_expire(struct mount *mnt)
956 mnt_add_count(mnt, -1);
957 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
962 if (mnt_get_count(mnt)) {
967 if (unlikely(mnt->mnt_pinned)) {
968 mnt_add_count(mnt, mnt->mnt_pinned + 1);
972 acct_auto_close_mnt(&mnt->mnt);
975 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
980 mnt->mnt.mnt_flags |= MNT_DOOMED;
983 list_del(&mnt->mnt_instance);
987 * This probably indicates that somebody messed
988 * up a mnt_want/drop_write() pair. If this
989 * happens, the filesystem was probably unable
990 * to make r/w->r/o transitions.
993 * The locking used to deal with mnt_count decrement provides barriers,
994 * so mnt_get_writers() below is safe.
996 WARN_ON(mnt_get_writers(mnt));
997 fsnotify_vfsmount_delete(&mnt->mnt);
998 dput(mnt->mnt.mnt_root);
999 deactivate_super(mnt->mnt.mnt_sb);
1001 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1004 void mntput(struct vfsmount *mnt)
1007 struct mount *m = real_mount(mnt);
1008 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1009 if (unlikely(m->mnt_expiry_mark))
1010 m->mnt_expiry_mark = 0;
1011 mntput_no_expire(m);
1014 EXPORT_SYMBOL(mntput);
1016 struct vfsmount *mntget(struct vfsmount *mnt)
1019 mnt_add_count(real_mount(mnt), 1);
1022 EXPORT_SYMBOL(mntget);
1024 void mnt_pin(struct vfsmount *mnt)
1027 real_mount(mnt)->mnt_pinned++;
1028 unlock_mount_hash();
1030 EXPORT_SYMBOL(mnt_pin);
1032 void mnt_unpin(struct vfsmount *m)
1034 struct mount *mnt = real_mount(m);
1036 if (mnt->mnt_pinned) {
1037 mnt_add_count(mnt, 1);
1040 unlock_mount_hash();
1042 EXPORT_SYMBOL(mnt_unpin);
1044 static inline void mangle(struct seq_file *m, const char *s)
1046 seq_escape(m, s, " \t\n\\");
1050 * Simple .show_options callback for filesystems which don't want to
1051 * implement more complex mount option showing.
1053 * See also save_mount_options().
1055 int generic_show_options(struct seq_file *m, struct dentry *root)
1057 const char *options;
1060 options = rcu_dereference(root->d_sb->s_options);
1062 if (options != NULL && options[0]) {
1070 EXPORT_SYMBOL(generic_show_options);
1073 * If filesystem uses generic_show_options(), this function should be
1074 * called from the fill_super() callback.
1076 * The .remount_fs callback usually needs to be handled in a special
1077 * way, to make sure, that previous options are not overwritten if the
1080 * Also note, that if the filesystem's .remount_fs function doesn't
1081 * reset all options to their default value, but changes only newly
1082 * given options, then the displayed options will not reflect reality
1085 void save_mount_options(struct super_block *sb, char *options)
1087 BUG_ON(sb->s_options);
1088 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1090 EXPORT_SYMBOL(save_mount_options);
1092 void replace_mount_options(struct super_block *sb, char *options)
1094 char *old = sb->s_options;
1095 rcu_assign_pointer(sb->s_options, options);
1101 EXPORT_SYMBOL(replace_mount_options);
1103 #ifdef CONFIG_PROC_FS
1104 /* iterator; we want it to have access to namespace_sem, thus here... */
1105 static void *m_start(struct seq_file *m, loff_t *pos)
1107 struct proc_mounts *p = proc_mounts(m);
1109 down_read(&namespace_sem);
1110 if (p->cached_event == p->ns->event) {
1111 void *v = p->cached_mount;
1112 if (*pos == p->cached_index)
1114 if (*pos == p->cached_index + 1) {
1115 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1116 return p->cached_mount = v;
1120 p->cached_event = p->ns->event;
1121 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1122 p->cached_index = *pos;
1123 return p->cached_mount;
1126 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1128 struct proc_mounts *p = proc_mounts(m);
1130 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1131 p->cached_index = *pos;
1132 return p->cached_mount;
1135 static void m_stop(struct seq_file *m, void *v)
1137 up_read(&namespace_sem);
1140 static int m_show(struct seq_file *m, void *v)
1142 struct proc_mounts *p = proc_mounts(m);
1143 struct mount *r = list_entry(v, struct mount, mnt_list);
1144 return p->show(m, &r->mnt);
1147 const struct seq_operations mounts_op = {
1153 #endif /* CONFIG_PROC_FS */
1156 * may_umount_tree - check if a mount tree is busy
1157 * @mnt: root of mount tree
1159 * This is called to check if a tree of mounts has any
1160 * open files, pwds, chroots or sub mounts that are
1163 int may_umount_tree(struct vfsmount *m)
1165 struct mount *mnt = real_mount(m);
1166 int actual_refs = 0;
1167 int minimum_refs = 0;
1171 /* write lock needed for mnt_get_count */
1173 for (p = mnt; p; p = next_mnt(p, mnt)) {
1174 actual_refs += mnt_get_count(p);
1177 unlock_mount_hash();
1179 if (actual_refs > minimum_refs)
1185 EXPORT_SYMBOL(may_umount_tree);
1188 * may_umount - check if a mount point is busy
1189 * @mnt: root of mount
1191 * This is called to check if a mount point has any
1192 * open files, pwds, chroots or sub mounts. If the
1193 * mount has sub mounts this will return busy
1194 * regardless of whether the sub mounts are busy.
1196 * Doesn't take quota and stuff into account. IOW, in some cases it will
1197 * give false negatives. The main reason why it's here is that we need
1198 * a non-destructive way to look for easily umountable filesystems.
1200 int may_umount(struct vfsmount *mnt)
1203 down_read(&namespace_sem);
1205 if (propagate_mount_busy(real_mount(mnt), 2))
1207 unlock_mount_hash();
1208 up_read(&namespace_sem);
1212 EXPORT_SYMBOL(may_umount);
1214 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1216 static void namespace_unlock(void)
1219 struct hlist_head head = unmounted;
1221 if (likely(hlist_empty(&head))) {
1222 up_write(&namespace_sem);
1226 head.first->pprev = &head.first;
1227 INIT_HLIST_HEAD(&unmounted);
1229 up_write(&namespace_sem);
1233 while (!hlist_empty(&head)) {
1234 mnt = hlist_entry(head.first, struct mount, mnt_hash);
1235 hlist_del_init(&mnt->mnt_hash);
1236 if (mnt->mnt_ex_mountpoint.mnt)
1237 path_put(&mnt->mnt_ex_mountpoint);
1242 static inline void namespace_lock(void)
1244 down_write(&namespace_sem);
1248 * mount_lock must be held
1249 * namespace_sem must be held for write
1250 * how = 0 => just this tree, don't propagate
1251 * how = 1 => propagate; we know that nobody else has reference to any victims
1252 * how = 2 => lazy umount
1254 void umount_tree(struct mount *mnt, int how)
1256 HLIST_HEAD(tmp_list);
1258 struct mount *last = NULL;
1260 for (p = mnt; p; p = next_mnt(p, mnt)) {
1261 hlist_del_init_rcu(&p->mnt_hash);
1262 hlist_add_head(&p->mnt_hash, &tmp_list);
1266 propagate_umount(&tmp_list);
1268 hlist_for_each_entry(p, &tmp_list, mnt_hash) {
1269 list_del_init(&p->mnt_expire);
1270 list_del_init(&p->mnt_list);
1271 __touch_mnt_namespace(p->mnt_ns);
1274 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1275 list_del_init(&p->mnt_child);
1276 if (mnt_has_parent(p)) {
1277 put_mountpoint(p->mnt_mp);
1278 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1279 p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1280 p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1281 p->mnt_mountpoint = p->mnt.mnt_root;
1285 change_mnt_propagation(p, MS_PRIVATE);
1289 last->mnt_hash.next = unmounted.first;
1290 unmounted.first = tmp_list.first;
1291 unmounted.first->pprev = &unmounted.first;
1295 static void shrink_submounts(struct mount *mnt);
1297 static int do_umount(struct mount *mnt, int flags)
1299 struct super_block *sb = mnt->mnt.mnt_sb;
1302 retval = security_sb_umount(&mnt->mnt, flags);
1307 * Allow userspace to request a mountpoint be expired rather than
1308 * unmounting unconditionally. Unmount only happens if:
1309 * (1) the mark is already set (the mark is cleared by mntput())
1310 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1312 if (flags & MNT_EXPIRE) {
1313 if (&mnt->mnt == current->fs->root.mnt ||
1314 flags & (MNT_FORCE | MNT_DETACH))
1318 * probably don't strictly need the lock here if we examined
1319 * all race cases, but it's a slowpath.
1322 if (mnt_get_count(mnt) != 2) {
1323 unlock_mount_hash();
1326 unlock_mount_hash();
1328 if (!xchg(&mnt->mnt_expiry_mark, 1))
1333 * If we may have to abort operations to get out of this
1334 * mount, and they will themselves hold resources we must
1335 * allow the fs to do things. In the Unix tradition of
1336 * 'Gee thats tricky lets do it in userspace' the umount_begin
1337 * might fail to complete on the first run through as other tasks
1338 * must return, and the like. Thats for the mount program to worry
1339 * about for the moment.
1342 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1343 sb->s_op->umount_begin(sb);
1347 * No sense to grab the lock for this test, but test itself looks
1348 * somewhat bogus. Suggestions for better replacement?
1349 * Ho-hum... In principle, we might treat that as umount + switch
1350 * to rootfs. GC would eventually take care of the old vfsmount.
1351 * Actually it makes sense, especially if rootfs would contain a
1352 * /reboot - static binary that would close all descriptors and
1353 * call reboot(9). Then init(8) could umount root and exec /reboot.
1355 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1357 * Special case for "unmounting" root ...
1358 * we just try to remount it readonly.
1360 down_write(&sb->s_umount);
1361 if (!(sb->s_flags & MS_RDONLY))
1362 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1363 up_write(&sb->s_umount);
1371 if (flags & MNT_DETACH) {
1372 if (!list_empty(&mnt->mnt_list))
1373 umount_tree(mnt, 2);
1376 shrink_submounts(mnt);
1378 if (!propagate_mount_busy(mnt, 2)) {
1379 if (!list_empty(&mnt->mnt_list))
1380 umount_tree(mnt, 1);
1384 unlock_mount_hash();
1390 * Is the caller allowed to modify his namespace?
1392 static inline bool may_mount(void)
1394 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1398 * Now umount can handle mount points as well as block devices.
1399 * This is important for filesystems which use unnamed block devices.
1401 * We now support a flag for forced unmount like the other 'big iron'
1402 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1405 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1410 int lookup_flags = 0;
1412 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1418 if (!(flags & UMOUNT_NOFOLLOW))
1419 lookup_flags |= LOOKUP_FOLLOW;
1421 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1424 mnt = real_mount(path.mnt);
1426 if (path.dentry != path.mnt->mnt_root)
1428 if (!check_mnt(mnt))
1430 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1433 retval = do_umount(mnt, flags);
1435 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1437 mntput_no_expire(mnt);
1442 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1445 * The 2.0 compatible umount. No flags.
1447 SYSCALL_DEFINE1(oldumount, char __user *, name)
1449 return sys_umount(name, 0);
1454 static bool is_mnt_ns_file(struct dentry *dentry)
1456 /* Is this a proxy for a mount namespace? */
1457 struct inode *inode = dentry->d_inode;
1460 if (!proc_ns_inode(inode))
1463 ei = get_proc_ns(inode);
1464 if (ei->ns_ops != &mntns_operations)
1470 static bool mnt_ns_loop(struct dentry *dentry)
1472 /* Could bind mounting the mount namespace inode cause a
1473 * mount namespace loop?
1475 struct mnt_namespace *mnt_ns;
1476 if (!is_mnt_ns_file(dentry))
1479 mnt_ns = get_proc_ns(dentry->d_inode)->ns;
1480 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1483 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1486 struct mount *res, *p, *q, *r, *parent;
1488 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1489 return ERR_PTR(-EINVAL);
1491 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1492 return ERR_PTR(-EINVAL);
1494 res = q = clone_mnt(mnt, dentry, flag);
1498 q->mnt.mnt_flags &= ~MNT_LOCKED;
1499 q->mnt_mountpoint = mnt->mnt_mountpoint;
1502 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1504 if (!is_subdir(r->mnt_mountpoint, dentry))
1507 for (s = r; s; s = next_mnt(s, r)) {
1508 if (!(flag & CL_COPY_UNBINDABLE) &&
1509 IS_MNT_UNBINDABLE(s)) {
1510 s = skip_mnt_tree(s);
1513 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1514 is_mnt_ns_file(s->mnt.mnt_root)) {
1515 s = skip_mnt_tree(s);
1518 while (p != s->mnt_parent) {
1524 q = clone_mnt(p, p->mnt.mnt_root, flag);
1528 list_add_tail(&q->mnt_list, &res->mnt_list);
1529 attach_mnt(q, parent, p->mnt_mp);
1530 unlock_mount_hash();
1537 umount_tree(res, 0);
1538 unlock_mount_hash();
1543 /* Caller should check returned pointer for errors */
1545 struct vfsmount *collect_mounts(struct path *path)
1549 tree = copy_tree(real_mount(path->mnt), path->dentry,
1550 CL_COPY_ALL | CL_PRIVATE);
1553 return ERR_CAST(tree);
1557 void drop_collected_mounts(struct vfsmount *mnt)
1561 umount_tree(real_mount(mnt), 0);
1562 unlock_mount_hash();
1566 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1567 struct vfsmount *root)
1570 int res = f(root, arg);
1573 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1574 res = f(&mnt->mnt, arg);
1581 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1585 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1586 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1587 mnt_release_group_id(p);
1591 static int invent_group_ids(struct mount *mnt, bool recurse)
1595 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1596 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1597 int err = mnt_alloc_group_id(p);
1599 cleanup_group_ids(mnt, p);
1609 * @source_mnt : mount tree to be attached
1610 * @nd : place the mount tree @source_mnt is attached
1611 * @parent_nd : if non-null, detach the source_mnt from its parent and
1612 * store the parent mount and mountpoint dentry.
1613 * (done when source_mnt is moved)
1615 * NOTE: in the table below explains the semantics when a source mount
1616 * of a given type is attached to a destination mount of a given type.
1617 * ---------------------------------------------------------------------------
1618 * | BIND MOUNT OPERATION |
1619 * |**************************************************************************
1620 * | source-->| shared | private | slave | unbindable |
1624 * |**************************************************************************
1625 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1627 * |non-shared| shared (+) | private | slave (*) | invalid |
1628 * ***************************************************************************
1629 * A bind operation clones the source mount and mounts the clone on the
1630 * destination mount.
1632 * (++) the cloned mount is propagated to all the mounts in the propagation
1633 * tree of the destination mount and the cloned mount is added to
1634 * the peer group of the source mount.
1635 * (+) the cloned mount is created under the destination mount and is marked
1636 * as shared. The cloned mount is added to the peer group of the source
1638 * (+++) the mount is propagated to all the mounts in the propagation tree
1639 * of the destination mount and the cloned mount is made slave
1640 * of the same master as that of the source mount. The cloned mount
1641 * is marked as 'shared and slave'.
1642 * (*) the cloned mount is made a slave of the same master as that of the
1645 * ---------------------------------------------------------------------------
1646 * | MOVE MOUNT OPERATION |
1647 * |**************************************************************************
1648 * | source-->| shared | private | slave | unbindable |
1652 * |**************************************************************************
1653 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1655 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1656 * ***************************************************************************
1658 * (+) the mount is moved to the destination. And is then propagated to
1659 * all the mounts in the propagation tree of the destination mount.
1660 * (+*) the mount is moved to the destination.
1661 * (+++) the mount is moved to the destination and is then propagated to
1662 * all the mounts belonging to the destination mount's propagation tree.
1663 * the mount is marked as 'shared and slave'.
1664 * (*) the mount continues to be a slave at the new location.
1666 * if the source mount is a tree, the operations explained above is
1667 * applied to each mount in the tree.
1668 * Must be called without spinlocks held, since this function can sleep
1671 static int attach_recursive_mnt(struct mount *source_mnt,
1672 struct mount *dest_mnt,
1673 struct mountpoint *dest_mp,
1674 struct path *parent_path)
1676 HLIST_HEAD(tree_list);
1677 struct mount *child, *p;
1678 struct hlist_node *n;
1681 if (IS_MNT_SHARED(dest_mnt)) {
1682 err = invent_group_ids(source_mnt, true);
1685 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1688 goto out_cleanup_ids;
1689 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1695 detach_mnt(source_mnt, parent_path);
1696 attach_mnt(source_mnt, dest_mnt, dest_mp);
1697 touch_mnt_namespace(source_mnt->mnt_ns);
1699 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1700 commit_tree(source_mnt, NULL);
1703 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1705 hlist_del_init(&child->mnt_hash);
1706 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1707 child->mnt_mountpoint);
1708 commit_tree(child, q);
1710 unlock_mount_hash();
1715 while (!hlist_empty(&tree_list)) {
1716 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1717 umount_tree(child, 0);
1719 unlock_mount_hash();
1720 cleanup_group_ids(source_mnt, NULL);
1725 static struct mountpoint *lock_mount(struct path *path)
1727 struct vfsmount *mnt;
1728 struct dentry *dentry = path->dentry;
1730 mutex_lock(&dentry->d_inode->i_mutex);
1731 if (unlikely(cant_mount(dentry))) {
1732 mutex_unlock(&dentry->d_inode->i_mutex);
1733 return ERR_PTR(-ENOENT);
1736 mnt = lookup_mnt(path);
1738 struct mountpoint *mp = new_mountpoint(dentry);
1741 mutex_unlock(&dentry->d_inode->i_mutex);
1747 mutex_unlock(&path->dentry->d_inode->i_mutex);
1750 dentry = path->dentry = dget(mnt->mnt_root);
1754 static void unlock_mount(struct mountpoint *where)
1756 struct dentry *dentry = where->m_dentry;
1757 put_mountpoint(where);
1759 mutex_unlock(&dentry->d_inode->i_mutex);
1762 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1764 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1767 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1768 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1771 return attach_recursive_mnt(mnt, p, mp, NULL);
1775 * Sanity check the flags to change_mnt_propagation.
1778 static int flags_to_propagation_type(int flags)
1780 int type = flags & ~(MS_REC | MS_SILENT);
1782 /* Fail if any non-propagation flags are set */
1783 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1785 /* Only one propagation flag should be set */
1786 if (!is_power_of_2(type))
1792 * recursively change the type of the mountpoint.
1794 static int do_change_type(struct path *path, int flag)
1797 struct mount *mnt = real_mount(path->mnt);
1798 int recurse = flag & MS_REC;
1802 if (path->dentry != path->mnt->mnt_root)
1805 type = flags_to_propagation_type(flag);
1810 if (type == MS_SHARED) {
1811 err = invent_group_ids(mnt, recurse);
1817 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1818 change_mnt_propagation(m, type);
1819 unlock_mount_hash();
1826 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1828 struct mount *child;
1829 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1830 if (!is_subdir(child->mnt_mountpoint, dentry))
1833 if (child->mnt.mnt_flags & MNT_LOCKED)
1840 * do loopback mount.
1842 static int do_loopback(struct path *path, const char *old_name,
1845 struct path old_path;
1846 struct mount *mnt = NULL, *old, *parent;
1847 struct mountpoint *mp;
1849 if (!old_name || !*old_name)
1851 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1856 if (mnt_ns_loop(old_path.dentry))
1859 mp = lock_mount(path);
1864 old = real_mount(old_path.mnt);
1865 parent = real_mount(path->mnt);
1868 if (IS_MNT_UNBINDABLE(old))
1871 if (!check_mnt(parent) || !check_mnt(old))
1874 if (!recurse && has_locked_children(old, old_path.dentry))
1878 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
1880 mnt = clone_mnt(old, old_path.dentry, 0);
1887 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1889 err = graft_tree(mnt, parent, mp);
1892 umount_tree(mnt, 0);
1893 unlock_mount_hash();
1898 path_put(&old_path);
1902 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1905 int readonly_request = 0;
1907 if (ms_flags & MS_RDONLY)
1908 readonly_request = 1;
1909 if (readonly_request == __mnt_is_readonly(mnt))
1912 if (readonly_request)
1913 error = mnt_make_readonly(real_mount(mnt));
1915 __mnt_unmake_readonly(real_mount(mnt));
1920 * change filesystem flags. dir should be a physical root of filesystem.
1921 * If you've mounted a non-root directory somewhere and want to do remount
1922 * on it - tough luck.
1924 static int do_remount(struct path *path, int flags, int mnt_flags,
1928 struct super_block *sb = path->mnt->mnt_sb;
1929 struct mount *mnt = real_mount(path->mnt);
1931 if (!check_mnt(mnt))
1934 if (path->dentry != path->mnt->mnt_root)
1937 /* Don't allow changing of locked mnt flags.
1939 * No locks need to be held here while testing the various
1940 * MNT_LOCK flags because those flags can never be cleared
1941 * once they are set.
1943 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
1944 !(mnt_flags & MNT_READONLY)) {
1947 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
1948 !(mnt_flags & MNT_NODEV)) {
1951 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
1952 !(mnt_flags & MNT_NOSUID)) {
1955 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
1956 !(mnt_flags & MNT_NOEXEC)) {
1959 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
1960 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
1964 err = security_sb_remount(sb, data);
1968 down_write(&sb->s_umount);
1969 if (flags & MS_BIND)
1970 err = change_mount_flags(path->mnt, flags);
1971 else if (!capable(CAP_SYS_ADMIN))
1974 err = do_remount_sb(sb, flags, data, 0);
1977 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
1978 mnt->mnt.mnt_flags = mnt_flags;
1979 touch_mnt_namespace(mnt->mnt_ns);
1980 unlock_mount_hash();
1982 up_write(&sb->s_umount);
1986 static inline int tree_contains_unbindable(struct mount *mnt)
1989 for (p = mnt; p; p = next_mnt(p, mnt)) {
1990 if (IS_MNT_UNBINDABLE(p))
1996 static int do_move_mount(struct path *path, const char *old_name)
1998 struct path old_path, parent_path;
2001 struct mountpoint *mp;
2003 if (!old_name || !*old_name)
2005 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2009 mp = lock_mount(path);
2014 old = real_mount(old_path.mnt);
2015 p = real_mount(path->mnt);
2018 if (!check_mnt(p) || !check_mnt(old))
2021 if (old->mnt.mnt_flags & MNT_LOCKED)
2025 if (old_path.dentry != old_path.mnt->mnt_root)
2028 if (!mnt_has_parent(old))
2031 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
2032 S_ISDIR(old_path.dentry->d_inode->i_mode))
2035 * Don't move a mount residing in a shared parent.
2037 if (IS_MNT_SHARED(old->mnt_parent))
2040 * Don't move a mount tree containing unbindable mounts to a destination
2041 * mount which is shared.
2043 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2046 for (; mnt_has_parent(p); p = p->mnt_parent)
2050 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2054 /* if the mount is moved, it should no longer be expire
2056 list_del_init(&old->mnt_expire);
2061 path_put(&parent_path);
2062 path_put(&old_path);
2066 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2069 const char *subtype = strchr(fstype, '.');
2078 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2080 if (!mnt->mnt_sb->s_subtype)
2086 return ERR_PTR(err);
2090 * add a mount into a namespace's mount tree
2092 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2094 struct mountpoint *mp;
2095 struct mount *parent;
2098 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2100 mp = lock_mount(path);
2104 parent = real_mount(path->mnt);
2106 if (unlikely(!check_mnt(parent))) {
2107 /* that's acceptable only for automounts done in private ns */
2108 if (!(mnt_flags & MNT_SHRINKABLE))
2110 /* ... and for those we'd better have mountpoint still alive */
2111 if (!parent->mnt_ns)
2115 /* Refuse the same filesystem on the same mount point */
2117 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2118 path->mnt->mnt_root == path->dentry)
2122 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2125 newmnt->mnt.mnt_flags = mnt_flags;
2126 err = graft_tree(newmnt, parent, mp);
2134 * create a new mount for userspace and request it to be added into the
2137 static int do_new_mount(struct path *path, const char *fstype, int flags,
2138 int mnt_flags, const char *name, void *data)
2140 struct file_system_type *type;
2141 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2142 struct vfsmount *mnt;
2148 type = get_fs_type(fstype);
2152 if (user_ns != &init_user_ns) {
2153 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2154 put_filesystem(type);
2157 /* Only in special cases allow devices from mounts
2158 * created outside the initial user namespace.
2160 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2162 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2166 mnt = vfs_kern_mount(type, flags, name, data);
2167 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2168 !mnt->mnt_sb->s_subtype)
2169 mnt = fs_set_subtype(mnt, fstype);
2171 put_filesystem(type);
2173 return PTR_ERR(mnt);
2175 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2181 int finish_automount(struct vfsmount *m, struct path *path)
2183 struct mount *mnt = real_mount(m);
2185 /* The new mount record should have at least 2 refs to prevent it being
2186 * expired before we get a chance to add it
2188 BUG_ON(mnt_get_count(mnt) < 2);
2190 if (m->mnt_sb == path->mnt->mnt_sb &&
2191 m->mnt_root == path->dentry) {
2196 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2200 /* remove m from any expiration list it may be on */
2201 if (!list_empty(&mnt->mnt_expire)) {
2203 list_del_init(&mnt->mnt_expire);
2212 * mnt_set_expiry - Put a mount on an expiration list
2213 * @mnt: The mount to list.
2214 * @expiry_list: The list to add the mount to.
2216 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2220 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2224 EXPORT_SYMBOL(mnt_set_expiry);
2227 * process a list of expirable mountpoints with the intent of discarding any
2228 * mountpoints that aren't in use and haven't been touched since last we came
2231 void mark_mounts_for_expiry(struct list_head *mounts)
2233 struct mount *mnt, *next;
2234 LIST_HEAD(graveyard);
2236 if (list_empty(mounts))
2242 /* extract from the expiration list every vfsmount that matches the
2243 * following criteria:
2244 * - only referenced by its parent vfsmount
2245 * - still marked for expiry (marked on the last call here; marks are
2246 * cleared by mntput())
2248 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2249 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2250 propagate_mount_busy(mnt, 1))
2252 list_move(&mnt->mnt_expire, &graveyard);
2254 while (!list_empty(&graveyard)) {
2255 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2256 touch_mnt_namespace(mnt->mnt_ns);
2257 umount_tree(mnt, 1);
2259 unlock_mount_hash();
2263 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2266 * Ripoff of 'select_parent()'
2268 * search the list of submounts for a given mountpoint, and move any
2269 * shrinkable submounts to the 'graveyard' list.
2271 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2273 struct mount *this_parent = parent;
2274 struct list_head *next;
2278 next = this_parent->mnt_mounts.next;
2280 while (next != &this_parent->mnt_mounts) {
2281 struct list_head *tmp = next;
2282 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2285 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2288 * Descend a level if the d_mounts list is non-empty.
2290 if (!list_empty(&mnt->mnt_mounts)) {
2295 if (!propagate_mount_busy(mnt, 1)) {
2296 list_move_tail(&mnt->mnt_expire, graveyard);
2301 * All done at this level ... ascend and resume the search
2303 if (this_parent != parent) {
2304 next = this_parent->mnt_child.next;
2305 this_parent = this_parent->mnt_parent;
2312 * process a list of expirable mountpoints with the intent of discarding any
2313 * submounts of a specific parent mountpoint
2315 * mount_lock must be held for write
2317 static void shrink_submounts(struct mount *mnt)
2319 LIST_HEAD(graveyard);
2322 /* extract submounts of 'mountpoint' from the expiration list */
2323 while (select_submounts(mnt, &graveyard)) {
2324 while (!list_empty(&graveyard)) {
2325 m = list_first_entry(&graveyard, struct mount,
2327 touch_mnt_namespace(m->mnt_ns);
2334 * Some copy_from_user() implementations do not return the exact number of
2335 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2336 * Note that this function differs from copy_from_user() in that it will oops
2337 * on bad values of `to', rather than returning a short copy.
2339 static long exact_copy_from_user(void *to, const void __user * from,
2343 const char __user *f = from;
2346 if (!access_ok(VERIFY_READ, from, n))
2350 if (__get_user(c, f)) {
2361 int copy_mount_options(const void __user * data, unsigned long *where)
2371 if (!(page = __get_free_page(GFP_KERNEL)))
2374 /* We only care that *some* data at the address the user
2375 * gave us is valid. Just in case, we'll zero
2376 * the remainder of the page.
2378 /* copy_from_user cannot cross TASK_SIZE ! */
2379 size = TASK_SIZE - (unsigned long)data;
2380 if (size > PAGE_SIZE)
2383 i = size - exact_copy_from_user((void *)page, data, size);
2389 memset((char *)page + i, 0, PAGE_SIZE - i);
2394 int copy_mount_string(const void __user *data, char **where)
2403 tmp = strndup_user(data, PAGE_SIZE);
2405 return PTR_ERR(tmp);
2412 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2413 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2415 * data is a (void *) that can point to any structure up to
2416 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2417 * information (or be NULL).
2419 * Pre-0.97 versions of mount() didn't have a flags word.
2420 * When the flags word was introduced its top half was required
2421 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2422 * Therefore, if this magic number is present, it carries no information
2423 * and must be discarded.
2425 long do_mount(const char *dev_name, const char *dir_name,
2426 const char *type_page, unsigned long flags, void *data_page)
2433 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2434 flags &= ~MS_MGC_MSK;
2436 /* Basic sanity checks */
2438 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2442 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2444 /* ... and get the mountpoint */
2445 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2449 retval = security_sb_mount(dev_name, &path,
2450 type_page, flags, data_page);
2451 if (!retval && !may_mount())
2456 /* Default to relatime unless overriden */
2457 if (!(flags & MS_NOATIME))
2458 mnt_flags |= MNT_RELATIME;
2460 /* Separate the per-mountpoint flags */
2461 if (flags & MS_NOSUID)
2462 mnt_flags |= MNT_NOSUID;
2463 if (flags & MS_NODEV)
2464 mnt_flags |= MNT_NODEV;
2465 if (flags & MS_NOEXEC)
2466 mnt_flags |= MNT_NOEXEC;
2467 if (flags & MS_NOATIME)
2468 mnt_flags |= MNT_NOATIME;
2469 if (flags & MS_NODIRATIME)
2470 mnt_flags |= MNT_NODIRATIME;
2471 if (flags & MS_STRICTATIME)
2472 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2473 if (flags & MS_RDONLY)
2474 mnt_flags |= MNT_READONLY;
2476 /* The default atime for remount is preservation */
2477 if ((flags & MS_REMOUNT) &&
2478 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2479 MS_STRICTATIME)) == 0)) {
2480 mnt_flags &= ~MNT_ATIME_MASK;
2481 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2484 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2485 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2488 if (flags & MS_REMOUNT)
2489 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2491 else if (flags & MS_BIND)
2492 retval = do_loopback(&path, dev_name, flags & MS_REC);
2493 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2494 retval = do_change_type(&path, flags);
2495 else if (flags & MS_MOVE)
2496 retval = do_move_mount(&path, dev_name);
2498 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2499 dev_name, data_page);
2505 static void free_mnt_ns(struct mnt_namespace *ns)
2507 proc_free_inum(ns->proc_inum);
2508 put_user_ns(ns->user_ns);
2513 * Assign a sequence number so we can detect when we attempt to bind
2514 * mount a reference to an older mount namespace into the current
2515 * mount namespace, preventing reference counting loops. A 64bit
2516 * number incrementing at 10Ghz will take 12,427 years to wrap which
2517 * is effectively never, so we can ignore the possibility.
2519 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2521 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2523 struct mnt_namespace *new_ns;
2526 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2528 return ERR_PTR(-ENOMEM);
2529 ret = proc_alloc_inum(&new_ns->proc_inum);
2532 return ERR_PTR(ret);
2534 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2535 atomic_set(&new_ns->count, 1);
2536 new_ns->root = NULL;
2537 INIT_LIST_HEAD(&new_ns->list);
2538 init_waitqueue_head(&new_ns->poll);
2540 new_ns->user_ns = get_user_ns(user_ns);
2544 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2545 struct user_namespace *user_ns, struct fs_struct *new_fs)
2547 struct mnt_namespace *new_ns;
2548 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2549 struct mount *p, *q;
2556 if (likely(!(flags & CLONE_NEWNS))) {
2563 new_ns = alloc_mnt_ns(user_ns);
2568 /* First pass: copy the tree topology */
2569 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2570 if (user_ns != ns->user_ns)
2571 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2572 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2575 free_mnt_ns(new_ns);
2576 return ERR_CAST(new);
2579 list_add_tail(&new_ns->list, &new->mnt_list);
2582 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2583 * as belonging to new namespace. We have already acquired a private
2584 * fs_struct, so tsk->fs->lock is not needed.
2591 if (&p->mnt == new_fs->root.mnt) {
2592 new_fs->root.mnt = mntget(&q->mnt);
2595 if (&p->mnt == new_fs->pwd.mnt) {
2596 new_fs->pwd.mnt = mntget(&q->mnt);
2600 p = next_mnt(p, old);
2601 q = next_mnt(q, new);
2604 while (p->mnt.mnt_root != q->mnt.mnt_root)
2605 p = next_mnt(p, old);
2618 * create_mnt_ns - creates a private namespace and adds a root filesystem
2619 * @mnt: pointer to the new root filesystem mountpoint
2621 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2623 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2624 if (!IS_ERR(new_ns)) {
2625 struct mount *mnt = real_mount(m);
2626 mnt->mnt_ns = new_ns;
2628 list_add(&mnt->mnt_list, &new_ns->list);
2635 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2637 struct mnt_namespace *ns;
2638 struct super_block *s;
2642 ns = create_mnt_ns(mnt);
2644 return ERR_CAST(ns);
2646 err = vfs_path_lookup(mnt->mnt_root, mnt,
2647 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2652 return ERR_PTR(err);
2654 /* trade a vfsmount reference for active sb one */
2655 s = path.mnt->mnt_sb;
2656 atomic_inc(&s->s_active);
2658 /* lock the sucker */
2659 down_write(&s->s_umount);
2660 /* ... and return the root of (sub)tree on it */
2663 EXPORT_SYMBOL(mount_subtree);
2665 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2666 char __user *, type, unsigned long, flags, void __user *, data)
2670 struct filename *kernel_dir;
2672 unsigned long data_page;
2674 ret = copy_mount_string(type, &kernel_type);
2678 kernel_dir = getname(dir_name);
2679 if (IS_ERR(kernel_dir)) {
2680 ret = PTR_ERR(kernel_dir);
2684 ret = copy_mount_string(dev_name, &kernel_dev);
2688 ret = copy_mount_options(data, &data_page);
2692 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2693 (void *) data_page);
2695 free_page(data_page);
2699 putname(kernel_dir);
2707 * Return true if path is reachable from root
2709 * namespace_sem or mount_lock is held
2711 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2712 const struct path *root)
2714 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2715 dentry = mnt->mnt_mountpoint;
2716 mnt = mnt->mnt_parent;
2718 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2721 int path_is_under(struct path *path1, struct path *path2)
2724 read_seqlock_excl(&mount_lock);
2725 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2726 read_sequnlock_excl(&mount_lock);
2729 EXPORT_SYMBOL(path_is_under);
2732 * pivot_root Semantics:
2733 * Moves the root file system of the current process to the directory put_old,
2734 * makes new_root as the new root file system of the current process, and sets
2735 * root/cwd of all processes which had them on the current root to new_root.
2738 * The new_root and put_old must be directories, and must not be on the
2739 * same file system as the current process root. The put_old must be
2740 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2741 * pointed to by put_old must yield the same directory as new_root. No other
2742 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2744 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2745 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2746 * in this situation.
2749 * - we don't move root/cwd if they are not at the root (reason: if something
2750 * cared enough to change them, it's probably wrong to force them elsewhere)
2751 * - it's okay to pick a root that isn't the root of a file system, e.g.
2752 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2753 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2756 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2757 const char __user *, put_old)
2759 struct path new, old, parent_path, root_parent, root;
2760 struct mount *new_mnt, *root_mnt, *old_mnt;
2761 struct mountpoint *old_mp, *root_mp;
2767 error = user_path_dir(new_root, &new);
2771 error = user_path_dir(put_old, &old);
2775 error = security_sb_pivotroot(&old, &new);
2779 get_fs_root(current->fs, &root);
2780 old_mp = lock_mount(&old);
2781 error = PTR_ERR(old_mp);
2786 new_mnt = real_mount(new.mnt);
2787 root_mnt = real_mount(root.mnt);
2788 old_mnt = real_mount(old.mnt);
2789 if (IS_MNT_SHARED(old_mnt) ||
2790 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2791 IS_MNT_SHARED(root_mnt->mnt_parent))
2793 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2795 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2798 if (d_unlinked(new.dentry))
2801 if (new_mnt == root_mnt || old_mnt == root_mnt)
2802 goto out4; /* loop, on the same file system */
2804 if (root.mnt->mnt_root != root.dentry)
2805 goto out4; /* not a mountpoint */
2806 if (!mnt_has_parent(root_mnt))
2807 goto out4; /* not attached */
2808 root_mp = root_mnt->mnt_mp;
2809 if (new.mnt->mnt_root != new.dentry)
2810 goto out4; /* not a mountpoint */
2811 if (!mnt_has_parent(new_mnt))
2812 goto out4; /* not attached */
2813 /* make sure we can reach put_old from new_root */
2814 if (!is_path_reachable(old_mnt, old.dentry, &new))
2816 root_mp->m_count++; /* pin it so it won't go away */
2818 detach_mnt(new_mnt, &parent_path);
2819 detach_mnt(root_mnt, &root_parent);
2820 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2821 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2822 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2824 /* mount old root on put_old */
2825 attach_mnt(root_mnt, old_mnt, old_mp);
2826 /* mount new_root on / */
2827 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2828 touch_mnt_namespace(current->nsproxy->mnt_ns);
2829 unlock_mount_hash();
2830 chroot_fs_refs(&root, &new);
2831 put_mountpoint(root_mp);
2834 unlock_mount(old_mp);
2836 path_put(&root_parent);
2837 path_put(&parent_path);
2849 static void __init init_mount_tree(void)
2851 struct vfsmount *mnt;
2852 struct mnt_namespace *ns;
2854 struct file_system_type *type;
2856 type = get_fs_type("rootfs");
2858 panic("Can't find rootfs type");
2859 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2860 put_filesystem(type);
2862 panic("Can't create rootfs");
2864 ns = create_mnt_ns(mnt);
2866 panic("Can't allocate initial namespace");
2868 init_task.nsproxy->mnt_ns = ns;
2872 root.dentry = mnt->mnt_root;
2874 set_fs_pwd(current->fs, &root);
2875 set_fs_root(current->fs, &root);
2878 void __init mnt_init(void)
2883 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2884 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2886 mount_hashtable = alloc_large_system_hash("Mount-cache",
2887 sizeof(struct hlist_head),
2890 &m_hash_shift, &m_hash_mask, 0, 0);
2891 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
2892 sizeof(struct hlist_head),
2895 &mp_hash_shift, &mp_hash_mask, 0, 0);
2897 if (!mount_hashtable || !mountpoint_hashtable)
2898 panic("Failed to allocate mount hash table\n");
2900 for (u = 0; u <= m_hash_mask; u++)
2901 INIT_HLIST_HEAD(&mount_hashtable[u]);
2902 for (u = 0; u <= mp_hash_mask; u++)
2903 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
2909 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2911 fs_kobj = kobject_create_and_add("fs", NULL);
2913 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2918 void put_mnt_ns(struct mnt_namespace *ns)
2920 if (!atomic_dec_and_test(&ns->count))
2922 drop_collected_mounts(&ns->root->mnt);
2926 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2928 struct vfsmount *mnt;
2929 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2932 * it is a longterm mount, don't release mnt until
2933 * we unmount before file sys is unregistered
2935 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2939 EXPORT_SYMBOL_GPL(kern_mount_data);
2941 void kern_unmount(struct vfsmount *mnt)
2943 /* release long term mount so mount point can be released */
2944 if (!IS_ERR_OR_NULL(mnt)) {
2945 real_mount(mnt)->mnt_ns = NULL;
2946 synchronize_rcu(); /* yecchhh... */
2950 EXPORT_SYMBOL(kern_unmount);
2952 bool our_mnt(struct vfsmount *mnt)
2954 return check_mnt(real_mount(mnt));
2957 bool current_chrooted(void)
2959 /* Does the current process have a non-standard root */
2960 struct path ns_root;
2961 struct path fs_root;
2964 /* Find the namespace root */
2965 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
2966 ns_root.dentry = ns_root.mnt->mnt_root;
2968 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2971 get_fs_root(current->fs, &fs_root);
2973 chrooted = !path_equal(&fs_root, &ns_root);
2981 bool fs_fully_visible(struct file_system_type *type)
2983 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2985 bool visible = false;
2990 down_read(&namespace_sem);
2991 list_for_each_entry(mnt, &ns->list, mnt_list) {
2992 struct mount *child;
2993 if (mnt->mnt.mnt_sb->s_type != type)
2996 /* This mount is not fully visible if there are any child mounts
2997 * that cover anything except for empty directories.
2999 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3000 struct inode *inode = child->mnt_mountpoint->d_inode;
3001 if (!S_ISDIR(inode->i_mode))
3003 if (inode->i_nlink > 2)
3011 up_read(&namespace_sem);
3015 static void *mntns_get(struct task_struct *task)
3017 struct mnt_namespace *ns = NULL;
3018 struct nsproxy *nsproxy;
3021 nsproxy = task->nsproxy;
3023 ns = nsproxy->mnt_ns;
3031 static void mntns_put(void *ns)
3036 static int mntns_install(struct nsproxy *nsproxy, void *ns)
3038 struct fs_struct *fs = current->fs;
3039 struct mnt_namespace *mnt_ns = ns;
3042 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3043 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3044 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3051 put_mnt_ns(nsproxy->mnt_ns);
3052 nsproxy->mnt_ns = mnt_ns;
3055 root.mnt = &mnt_ns->root->mnt;
3056 root.dentry = mnt_ns->root->mnt.mnt_root;
3058 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3061 /* Update the pwd and root */
3062 set_fs_pwd(fs, &root);
3063 set_fs_root(fs, &root);
3069 static unsigned int mntns_inum(void *ns)
3071 struct mnt_namespace *mnt_ns = ns;
3072 return mnt_ns->proc_inum;
3075 const struct proc_ns_operations mntns_operations = {
3077 .type = CLONE_NEWNS,
3080 .install = mntns_install,